This post is part of a series titled “OEDP’s Fieldnotes on AI”, where we offer reflections about AI in environmental and participatory contexts. This week’s reflections center on the question:
What is a point of view, and why can’t AI have one?
We look at what a “point of view” is in environmental research and EJ work—how POV in environmental data work is shaped by lived experience, institutional position, and power, and how it emerges from place, exposure, and lived environmental experience.
Environmental justice scholars remind us that justice is not singular or universal. It is multivalent. It includes power, recognition, relationality, and the ways communities understand land, harm, and responsibility. Justice frameworks are situated, emerging from particular histories, places, and epistemic traditions. What counts as harm, who counts as affected, and how “healthy” a community is are never neutral questions.
This insight feels particularly urgent as AI systems increasingly function as institutional actors. Agentic AI systems are not just tools; they shape meaning, distribute resources, and set the conditions in which both human and machine agency operate. They operate at machine speed, far faster than traditional environmental oversight cycles that can take years. Governance conversations focus on distributed (i.e., polycentric) models. Yet often missing in these conversations is the recognition that justice and decision-making are not abstract procedural matters, but grounded in lived exposure and what results.
I’m imagining any one of the people that I’ve (physically) stood next to in our shared work on environmental pollution. Many are women, many are elders, many have a sharp eye for seeing patterns in their community because they hold the pulse of their families. Which means they are also the ones who can see, and point to, the potential link between the nearby petrochemical refining facility and the rising cases of asthma or scleroderma among friends and family. They see how oil and gas companies are tied to power, and they live every day with the realities of being next to those facilities. This is their lived experience—not an opinion, but a demonstration of the way that point of view is shaped by our environments. And this point of view is one that will be carried into meetings and onto negotiating tables with company officials or regulatory and enforcement agencies. It will also be carried to the bedsides of unwell family members.
This is what a point of view is in environmental justice work. It’s not a preference, but a position in a landscape of exposure, risk, and consequence. It is formed through proximity to harm and proximity to power, and above all else, it is accountable because those who hold it must live with what happens next.
AI cannot have that kind of point of view. It can synthesize patterns across data and be trained on regulatory language, community testimony, epidemiological studies, and corporate reports. It can even structure decision-making, functioning as an institutional actor that influences distribution and interpretation. As Almeida, Filgueiras, and Mendonça argue in “Governing AI Agents with Democratic ‘Algorithmic Institutions,’” so-called agentic AI systems increasingly function not merely as tools but as constitutive institutions that structure interaction and shape contexts of action, raising questions about transparency, accountability, and human control. But they do not stand anywhere within the terrain they shape. They do not carry risk into a hospital room. They do not negotiate while fearing for a child’s ability to breathe. They do not live next to the refinery (maybe its physical manifestation as a data center does, but you get the point).
And this is where claims of neutrality become dangerous. When AI systems are framed as objective arbiters, they can obscure whose perspectives are encoded and whose are erased. In a community living next to a refinery, this can mean flattening lived patterns of illness into data points, while muting the people who first named the harm. Algorithms may operate at scale and speed, but they inherit the ways of knowing, and the blind spots, of the worlds that produced them. If dominant frameworks of justice already marginalize certain lived realities, AI systems trained within those frameworks risk reinforcing that marginalization, all while appearing neutral.
The governance conversation around agentic AI rightly calls for democratic design principles, human-led auditability, and preservation of moral agency. But environmental justice pushes that further. It asks: Which humans? Whose moral worlds? Whose exposure defines urgency? A distributed model of governance falls short if it stays abstracted from those who bear the brunt of environmental harm.
I find myself wondering, what can we build into these frameworks if AI cannot hold a point of view, but is nonethelss beginning to shape the landscape? What are the questions that we should be asking that help to strip away claims of neutrality and center multivalent perspectives and worldviews?
For our work with DIGITCORE, this might mean doubling-down on embedding provenance and positionality into data stewardship practices: Who gathered this data? Under what conditions? Who funded it? Who interprets it? Who bears the harm or benefit of its use? (For the record, this is work we should already be doing in participatory science.) It might also prompt us to consider what the design of AI-assisted tools that foreground uncertainty, dissent, and contextual narratives rather than collapsing them into a single output, can look like. It certainly means ensuring that community stewards, those with lived exposure, retain interpretive authority, even when AI systems assist with analysis.
If algorithms are [becoming] institutions, then we must ensure that they do not displace the experience and positions of those who live with consequence. Environmental justice teaches us that perspective is not noise, but the ground from which accountability emerges. AI may begin to shape the terrain of environmental policy and decision-making, but it cannot replace the situated human point of view that gives justice its meaning.
